Performance is an important issue for many-core architectures supported by networks-on-chip. One alternative for improving it is to map processes onto cores so as to mitigate the overall interprocess communication cost. In this context, this paper proposes the Kmeans algorithm as an alternative strategy to DRB and Greedy heuristics. For some communication patterns, our results pointed that Kmeans performs better than the other heuristics, thus being a good option for mapping processes on many-core architectures with networks-on-chip support.