A modular dynamic subgrid-scale modeling framework is presented for large eddy simulation of two-dimensional Boussinesq turbulence. The procedure we put forth in this study allows us to couple the structural subgrid-scale parameterization models with the functional models by minimizing the error between them. In particular, the approximate deconvolution procedure is used to estimate the Smagorinsky and Baldwin-Lomax eddy viscosity constants and the associated turbulent Prandtl numbers self-adaptively from the resolved flow quantities. Our numerical assessments for solving the Rayleigh-Benard turbulent thermal convection problem show that the proposed approach could be used as a viable tool to address the turbulence closure problem for the Boussinesq setting due to its accuracy and flexibility. (C) 2017 Elsevier Ltd. All rights reserved.