Methods of implementing an input-constrained, nonlinear, model-predictive controller in latent spaces using partial-least-squares (PLS)-based Hammerstein and Wiener models are discussed. For multiple-input, multiple-output (MIMO) systems, the PLS framework presents a viable alternative for identification and controller synthesis using Hammerstein and Wiener structures. The constraint mappings, which have to be taken into account during controller design in the PLS framework are highlighted PLS-based Wiener models are well suited for constrained control of nonlinear systems. The use of PLS-based Hammerstein models for control involves solution of a nonlinear program as a result of the constraint mapping. The proposed approach is demonstrated on a simulated pH-level control of an acid - base neutralization process.