The importance of developing predictive modeling tools has considerably increased with the diffusion of nanoscale technologies, in which strained layers and heterojunctions determine the materials modeling complexity. A self--consistent Poisson--Schroedinger solver based on a full-band k.p method within the envelope-function approximation has been developed and validated on large test structures, studying the oxide capacitance for various gate stacks. Among the effects governing the electrostatics of the devices we have studied (i) the impact of band structure models of the oxide, (ii) of the semiconductor, and (iii) the impact of the Si/SiO2 interfacial layers. The predictions of 2 advanced full band (FB) models (tight binding model and 30-bands k.p model) are compared to simpler 6-band k.p and 3-band effective mass approximation. In FB models, the oxides have been treated as pseudo zinc-blende materials, adjusting the model parameters using ab-initio simulations to match the band structure obtained from first principles. The importance of having such accurately calibrated models relies in the reduced computational time of EMA models with respect to FB models. Additionally we performed 2D TCAD simulations with a commercial simulation package to assess the predictions of the models implemented in TCAD tools. Finally we assessed the importance of accurate electrostatic modeling on gate tunneling current.