Modelling the interaction between flame and turbulence under the resolved scale depends on the closure reasonability. Physically, the flamelet assumptions remain reasonable in typical combustion simulations. Therefore species mass fractions and temperature can be mapped onto the mixture fraction, scalar dissipation and other relevant quantities. However, critical modeling closure problems do exist, especially the assumed joint probability density functions (PDFs) to calculate the filtered or mean field quantities. Aiming for such challenge issues, a newly proposed filtered turbulent flame model [1] is implemented in the framework of large eddy simulations for non-premixed turbulent combustion. The general principles of model realization are developed, including the structure and physics of the filtered flame, tabulation setup, and the numerical algorithm. Because of the involved modeling parameters for tabulation can be directly calculated from either the resolved or subgrid model quantities, the filtered turbulent flame model is in principle favorable to improve the numerical predictability. Tentative numerical tests and comparisons with other models justify this modeling idea.