You don't have to do anything - the compiler will do it all for you. The banks are contiguous in the memory map so large mallocs work, as well as declaring large arrays. You won't be able to malloc the whole 96Kb because Arduino uses some for itself. The largest I've practically needed so far was a 77Kb array char fb[240][320]; which worked with no problems at all. (You only need to worry about banks when dealing with very advanced DMA timing using multiple channels simultaneously)