In this work we introduce a new methodology to build a kernel matrix from a collection of kernels. The key idea is to build an unique kernel that eliminates spurious differences between kernels. We propose a method based on the Procrustes problems that uses the Alternating Projections method to minimize a certain error measure. The resulting kernel will be used for classification purposes using Support Vector Machines (SVMs). The proposed method has been successfully evaluated against alternative kernel combination techniques.