Fast Kernel Approximations for Latent Force Models and Convolved Multiple-Output Gaussian processes