Whatever the complexity - there is a way of running code on the Videocore GPU from the ARM CPU - and one of the routes to doing that is using an API called MMAL. GPU deinterlacing (which some may call 'hardware' deinterlacing - though in reality it's code running on the GPU rather than the CPU - which is how accelerated deinterelacing has improved over the years on the Pi series) still takes the workload away from the CPU (rather than what would be considered 'CPU' or 'software' deinterlacing where the CPU handles the code entirely.
I think it's tricky even to classify the Videocore as a GPU, it's more like an FPGA or a programmable DSP.
It's amazing we got 'software' decoding of HEVC on the RPi3, and that was on an older Videocore that had no business handling that logic.