Kernels can be seen as histogram generalizations. The kernel essentially represent different shapes of binning, shown below in the case of a 1D density distribution.