Memory operations on a custom backend

hardware-backends
5

Would it be possible to get a summary of the steps needed/to consider for integrating a custom accelerator aside from the kernel registrations? Or maybe that PR already contains everything?