I have read many times that random effects are not parameters of a linear mixed effects model but instead can be derived from the estimated variance/covariance parameters.

Can someone give an intuitive explanation how the variance/covariance parameters of the random effects can be estimated without actually using the random effects?