In this work, a novel data-driven design optimization framework for sizing and shaping topology optimization (TO) concepts efficiently by using size blending of implicit surfaces, free form deformation (FFD) shape morphing of meshes and multi-fidelity based ensemble of metamodels is developed. Instead of using standard CAD representations of the TO concepts, implicit surface based geometry (ISG) is utilized. The ISG is derived by adopting regularized radial basis function networks with compactly supported kernels. The sizing of the TO concepts are then generated as convex combinations of local blends of the ISG. A STL mesh of the sized ISG is generated using a marching cube algorithm, then FFD shape morphing is applied by putting anchor nodes on the mesh with prescribed deformations. Low and high fidelity computational experiments of the sized and shaped meshes using nonlinear finite element analysis are performed. The low fidelity data is generated using Hammersley sampling and the high fidelity data points are computed for optimal space filling designs of experiments. Finally, multi-fidelity based ensemble of metamodels are established for the simulated data and detailed size and shape optimization of the TO concept is performed simultaneously. The developed data-driven framework is efficient and robust. This is demonstrated for 3D benchmarks as well as a real engineering application of a casted mass eccentric flywheel for a compactor machine.