LLVM analysis on CUDA host code.

Is it possible to insert LLVM passes in CUDA host code. I want to insert some profiling and analysis passes on the host code but could not find any documentaion about it. If I understand correctly from the documentaion (https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#cuda-compilation-trajectory) default c++ compiler of the system is used for compiling the host code. Is it possible to modify the code during the process.