I don't know for sure, but I imagine that AROS's implementation of CGX is a thin layer around the HAL and that perhaps you'd end up having to recreate the latter, which may be quite difficult.
yes AROS use a HAL it is written in OO and i think its not very fast and do slowdown on slow systems.
the best way is to support the P96 driver interface and write a wrapper so for classic or winuae the P96 drivers can use.