Multiple kernel learning is a model to merge multiple kernels by linear combination. Mostly solving the models are slow due to explicit computation of kernels. Here, we propose to approximate kernel map function explicitly in finite dimensional sp