A comprehensive and accurate model for the prediction of critical heat flux (CHF) has been the talking point for improving the design and operational safety of high-power equipment like nuclear reactors. The presence of different CHF mechanisms over different flow regimes also adds to that difficulty. Machine Learning (ML) has shown potential for CHF prediction but limited by its data-driven knowledge, termed as black box with minimal interpretability. To address the limitation, a promising alternative physics-informed machine learning (PIML) appears that links data-driven insights with existing knowledge from a physical model. However, conventional PIML risks the possibility of results biasing towards a particular CHF type depending on the chosen physical model as the physical model is basically developed for a particular type of CHF mechanism. In this paper, we propose a new methodology by adapting multi-physical models for structuring the PIML for CHF prediction. Our method, termed multi-physics-informed machine learning (MPIML), integrates multiple physical models, each tailored to a distinct CHF mechanism. In this work, Lienhard correlation, Liu model and Katto model have been chosen to represent pool boiling, departure from nucleate boiling (DNB) in subcooled and low-quality regions, and liquid film dry-out in high-quality (annular flow) conditions for the prior knowledge of ML. This incorporation results in mitigating bias in the CHF prediction. The results demonstrate that MPIML significantly improves upon PIML approaches, enhancing both prediction accuracy and applicability across diverse CHF mechanisms.