We present a study based on a sample of 62 early-type galaxies (ETGs) at 0.9 < z spec < 2 aimed at constraining their past star formation and mass assembly histories. The sample is composed of normal ETGs having effective radii comparable to the mean radius of local ones and of compact ETGs having effective radii from two to six times smaller. We do not find evidence of a dependence of the compactness of ETGs on their stellar mass. The best fit to their spectral energy distribution at known redshift has allowed us to constrain the epoch at which the stellar mass formed. We find that the stellar mass of normal ETGs formed at z form≲ 3, while the stellar content of compact ETGs formed over a wider range of redshift (2 < z form < 10) with a large fraction of them characterized by z form > 5. Earlier stars, those formed at z form > 5, are assembled in compact and more massive ( M⊙) ETGs, while stars formed later (z form≲ 3) or resulting from subsequent episodes of star formation are assembled both in compact and in normal ETGs. Thus, the older the stellar population, the higher the mass of the hosting galaxy but not vice versa. This suggests that the epoch of formation may play a role in the formation of massive ETGs rather than the mass itself. We show that the possible general scheme in which normal ETGs at 〈z〉≃ 1.5 are descendants of compact spheroids assembled at higher redshift is not compatible with the current models. Indeed, we find that the number of dry mergers expected in a hierarchical model is almost two orders of magnitude lower than that needed to enlarge a compact ETG up to a normal-size ETG. Moreover, we do not find evidence supporting a dependence of the compactness of galaxies on their redshift of assembly, a dependence expected in the hypothesis that the compactness of a galaxy is due to the higher density of the Universe at earlier epochs. Finally, we propose a simple scheme of formation and assembly of the stellar mass of ETGs based on dissipative gas-rich merger, which can qualitatively account for the coexistence of normal and compact ETGs observed at 〈z〉≃ 1.5 in spite of the same stellar mass, the lack of normal ETGs with high z form and the absence of correlation between compactness, stellar mass and formation redshift.