Predictive control algorithms require an accurate mathematical model of the process under control. This paper shows how the classical Least-Squares algorithm can be extended to achieve simultaneous estimation of the parameters for all linear models with orders from 1 to n together with their associated loss functions in the delta-domain via a Lagrange multiplier technique. The resulting estimation algorithm can then be used to obtain adaptive, continuous-time and delta-domain Generalized Predictive Control algorithms.