This paper is devoted to the Dynamic Memory Allocation Problem (DMAP) in embedded systems. The existing Integer Linear Programing (ILP) formulation for DMAP is improved, and given that there are several metaheuristic approaches for solving the DMAP, a new metaheuristic approach is proposed and compared with the former ones. Computational results show that our new heuristic approach outperforms the best algorithm found in the literature regarding quality and running times.