In kernel density estimation, each observed data point is spread out by a "kernel" function such as a Gaussian, essentially exp(-0.5*x^2), or uniform function, f(x) = 1 for |x| < 0.5. So I think of a "kernel" as a function used to distribute a point mass.
