Finite element simulations constitute a key methodology for evaluating the reliability of electronic devices subjected to thermomechanical loading. In this work, a Krylov subspace-based model order reduction technique is applied to construct superelements (SEs), which are coupled with full-order finite element models (FEMs) within a mixed SE-FEM framework. Two representative electronic packaging assemblies - a flip-chip chip-scale package and a chip resistor - are systematically analyzed to quantify the influence of the number of interface nodes, governed by mesh resolution and element order, on the reduced model size, superelement generation cost, and overall computational efficiency.While speed-ups of up to 8.5 times are achieved for suitably chosen interface resolutions, excessively fine meshes or high-order interface discretizations may negate these efficiency gains and can even lead to increased computational cost. The SE-FEM strategy reproduces the original thermal response exactly and exhibits only weak sensitivity to the interface discretization. In contrast, the mechanical response shows a stronger dependence on interface resolution and mesh compatibility. These results underscore the necessity of carefully balancing interface mesh size and element order to achieve an optimal trade-off between computational efficiency and mechanical response accuracy. Furthermore, by eliminating internal degrees of freedom, the proposed SE framework provides an effective abstraction layer that facilitates model exchange while preserving intellectual property (IP).