Hyper-heuristics have emerged as effective general methodologies that are motivated by the goal of building or selecting heuristics automatically to solve a range of hard computational search problems with less development cost. HyFlex is a publicly available hyper-heuristic tool for rapid development and research which currently provides an interface to four problem domains along with relevant low level heuristics. A multistage hyper-heuristic based on random gradient and greedy with dominance heuristic selection methods is introduced in this study. This hyper-heuristic is implemented as an extension to HyFlex. The empirical results show that our approach performs better than some previously proposed hyper-heuristics over the given problem domains.