```python
import torch
import torch.nn as nn
from torch.utils.cpp_extension import load_inline

_cpp_source = r"""
#include <torch/extension.h>

torch::Tensor fused_mish_hardtanh_cuda(
    torch::Tensor input, double add_value, double scale);
"""

_cuda_source = r"""
#include <torch/extension.h>
#include <ATen/OpMathType.h>
#include <ATen/cuda/CUDAContext.h>
#include <c10/cuda/CUDAGuard.h>
#include <c10/cuda/CUDAException.h>
#include <cuda_runtime.h>
#include <algorithm>

template <typename T>
__device__ __forceinline__ T transform(T x, T add_value, T scale) {
    T softplus = x > T(20) ? x : log1p(exp(x));
    T y = x * tanh(softplus) + add_value;
    y = y < T(-1) ? T(-1) : y;
    y = y > T(1) ? T(1) : y;
    return y * scale;
}

template <>
__device__ __forceinline__ float transform<float>(
    float x, float add_value, float scale) {
    float softplus = x > 20.0f ? x : log1pf(expf(x));
    float y = x * tanhf(softplus) + add_value;
    y = y < -1.0f ? -1.0f : y;
    y = y > 1.0f ? 1.0f : y;
    return y * scale;
}

__global__ void fused_float4_kernel(
    const float4* __restrict__ input,
    float4* __restrict__ output,
    int64_t count,
    float add_value,
    float scale) {
    for (int64_t i = int64_t(blockIdx.x) * blockDim.x + threadIdx.x;
         i < count;
         i += int64_t(blockDim.x) * gridDim.x) {
        float4 v = input[i];
        v.x = transform(v.x, add_value, scale);
        v.y = transform(v.y, add_value, scale);
        v.z = transform(v.z, add_value, scale);
        v.w = transform(v.w, add_value, scale);
        output[i] = v;
    }
}

template <typename scalar_t>
__global__ void fused_scalar_kernel(
    const scalar_t* __restrict__ input,
    scalar_t* __restrict__ output,
    int64_t count,
    double add_value,
    double scale) {
    using acc_t = at::opmath_type<scalar_t>;
    for (int64_t i = int64_t(blockIdx.x) * blockDim.x + threadIdx.x;
         i < count;
         i += int64_t(blockDim.x) * gridDim.x) {
        acc_t x = static_cast<acc_t>(input[i]);
        acc_t sp = x > acc_t(20) ? x : log1p(exp(x));

        // Preserve intermediate rounding for half and bfloat16 tensors.
        scalar_t mish = static_cast<scalar_t>(x * tanh(sp));
        scalar_t added = static_cast<scalar_t>(
            static_cast<acc_t>(mish) + static_cast<acc_t>(add_value));
        acc_t y = static_cast<acc_t>(added);
        y = y < acc_t(-1) ? acc_t(-1) : y;
        y = y > acc_t(1) ? acc_t(1) : y;
        output[i] = static_cast<scalar_t>(y * static_cast<acc_t>(scale));
    }
}

torch::Tensor fused_mish_hardtanh_cuda(
    torch::Tensor input, double add_value, double scale) {
    TORCH_CHECK(input.is_cuda(), "input must be a CUDA tensor");
    c10::cuda::CUDAGuard guard(input.device());

    auto x = input.contiguous();
    auto output = torch::empty_like(x);
    int64_t count = x.numel();
    if (count == 0) {
        return output;
    }

    constexpr int threads = 256;
    auto stream = at::cuda::getCurrentCUDAStream();

    if (x.scalar_type() == at::kFloat && count % 4 == 0 &&
        reinterpret_cast<uintptr_t>(x.data_ptr<float>()) % 16 == 0) {
        int64_t vectors = count / 4;
        int blocks = static_cast<int>(
            std::min<int64_t>((vectors + threads - 1) / threads, 65535));

        fused_float4_kernel<<<blocks, threads, 0, stream>>>(
            reinterpret_cast<const float4*>(x.data_ptr<float>()),
            reinterpret_cast<float4*>(output.data_ptr<float>()),
            vectors, static_cast<float>(add_value), static_cast<float>(scale));
    } else {
        int blocks = static_cast<int>(
            std::min<int64_t>((count + threads - 1) / threads, 65535));

        AT_DISPATCH_FLOATING_TYPES_AND2(
            at::ScalarType::Half, at::ScalarType::BFloat16,
            x.scalar_type(), "fused_mish_hardtanh_cuda", [&] {
                fused_scalar_kernel<scalar_t><<<blocks, threads, 0, stream>>>(
                    x.data_ptr<scalar_t>(), output.data_ptr<scalar_t>(),
                    count, add_value, scale);
            });
    }

    C10_CUDA_KERNEL_LAUNCH_CHECK();
    return output;
}
"""

_fused_ops = load_inline(
    name="fused_mish_add_hardtanh_scale_v1",
    cpp_sources=_cpp_source,
    cuda_sources=_cuda_source,
    functions=["fused_mish_hardtanh_cuda"],
    extra_cflags=["-O3"],
    extra_cuda_cflags=["-O3"],
    verbose=False,
)


class ModelNew(nn.Module):
    def __init__(
        self,
        in_channels,
        out_channels,
        kernel_size,
        stride,
        padding,
        output_padding,
        add_value,
        scale,
    ):
        super().__init__()
        self.conv_transpose = nn.ConvTranspose2d(
            in_channels,
            out_channels,
            kernel_size,
            stride,
            padding,
            output_padding,
        )
        self.add_value = add_value
        self.scale = scale

    def forward(self, x):
        x = self.conv_transpose(x)
        return _fused_ops.fused_mish_hardtanh_cuda(
            x, float(self.add_value), float(self.scale)
        )
```