The article introduces a new way of using kernel machines with two layers, which is a type of computational method. This approach connects complex computational architectures with the concept of kernel learning. The researchers show that combining different types of kernels in two layers can lead to optimal results in solving problems. They also demonstrate that multiple kernel learning methods are a specific case of this two-layer approach. A new method called RLS2 is introduced, which performs well on various learning tasks. An open-source MATLAB toolbox is available for training and validating models using this method.