In this paper, we aim to improve the overall performance of kernel adaptive filters by adaptively combining several component filters with different parameters setting in the practical applications. The convex combination scheme is exploited to incorporate any two parallel diversity branches which could be the component filter or the output of previous combination layer. The proposed convex combination of multiple kernel adaptive filters can provide the more robust and better performance than the single filters with fixed parameters especially in the nonstationary complex environments without priori knowledge. Simulation results illustrate the superior performance of the proposed approach.