The question is related to the link : https://en.wikipedia.org/wiki/Reproducing_kernel_Hilbert_space
As I understand the Mercer theorem we can get RKHS. This lead to the kernel trick as mentioned in the link: https://en.wikipedia.org/wiki/Kernel_method
The above discussion leads to the idea that if we have vector (as a feature in machine learning) we can evaluate the kernel and get a corresponding scalar value. This is clear that in Kernel trick we are dealing with vectors.
The question is can this be extended to matrix as an input, instead of vectors. Any comments would be appreciated.