This project aims at investigating the potential of several memory optimizations on a variety of architectures. Target optimizations will leverage topology properties of systems to investigate data and threads allocation and migrations strategies. Furthermore, this project will investigate cross device data movements and hardware engines to offload host or accelerator burden. Viable optimizations will be integrated into AML library. AML is a memory management library developed at Argonne National Laboratory. The library provides customizable base abstractions as well as application level optimizations for memory mapping and data movements.