In this talk, I will give a brief introduction to learning with kernels, a fundamental area in machine learning. I will then present an overview of two related publications from my work. The first focuses on multiple kernel learning (MKL), which combines multiple kernel matrices and a domain related information source in an optimization model. I will discuss how this method can perform well on high dimensional and highly correlated data, how it can achieve interpretability, and how it can yield improved outcomes using only one tenth of the data. If time allows, I will also address a well-known limitation of kernel methods, namely the high computational cost, and introduce Multiple Approximate Kernel Learning, a novel approach that provides scalability and efficiency by using approximated matrices in place of full kernel matrices in MKL.