Now, OpenCV has code for old compute capability.

• https://github.com/opencv/opencv/blob/4.2.0/modules/core/include/opencv2/core/cuda/simd_functions.hpp#L95-L103
• https://github.com/opencv/opencv/blob/4.2.0/modules/core/include/opencv2/core/cuda/saturate_cast.hpp#L105-L107
• https://github.com/opencv/opencv/blob/4.2.0/modules/core/include/opencv2/core/cuda/scan.hpp#L196-L207

But, the latest CUDA Toolkit will stop to support old compute capability.
https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#deprecated-features

• Support for the following compute capabilities are deprecated in CUDA 10.2. Note that support for these compute capabilities may be removed in a future release of CUDA.
  • sm_35 (Kepler)
  • sm_37 (Kepler)
  • sm_50 (Maxwell)

So, I think that OpenCV had better to remove code for old compute capability.
It's also effective in terms of maintenance.