Yes, it is possible. The matrix to use would be a beta matrix. If a latent factor V is defined by other latent variables U_1 to U_p, a (p by 1) beta matrix with one of the paths fixed to 1.0 will do the trick.
Simplest case is the ‘hierarchical random-intercept’:
Here is an example of random-slopes for a latent DV.
In each case, a factor is defined by other latent factors.