Have you tried logarithms?
I’m not sure if this is a bad math joke about sleeping like a log, or if this is an actual suggestion for dealing with Gaussian process convolutions.
I’ve spent the last hour lying here trying to work out what the log of a GP-convolution would be.