The plasmon enhancement of molecular hyper-Raman scattering, the nonlinear counterpart of Raman scattering, which involves the absorption of two fundamental photons, is investigated with emphasis on low energy molecular vibrations. The two-photon excitation of the molecule is treated using its hyperpolarizability beta, and the emission of the hyper-Raman photons is computed using a dipole emitter located at the molecule position. The electromagnetic response of the plasmonic systems is evaluated using a surface integral equation method, which makes possible considering both planewave and dipole excitations in a single formalism. Taking into account different geometries (including multiresonant antennas and silver heptamers supporting Fano resonances), the experimental parameters influencing the enhancement of the molecular hyper-Raman scattering are discussed in detail. In particular, it is shown that a good excitation at the fundamental stage is not sufficient for reaching a good enhancement factor and that an optimization of the electromagnetic response of the plasmonic substrate is also important at the emission wavelength. The competition between the molecular hyper-Raman scattering signal and the background signal, that is, the second harmonic generation, is discussed. The latter can be reduced in specific structures by taking advantages of the key role played by the symmetry of the structure for hyper-Raman scattering and second harmonic generation. This way, we propose a nanostructure where the second harmonic generation can be reduced in the detection direction, enabling the hyper-Raman scattering signal from single donor-acceptor "push-pull" chromophores to be experimentally recorded with a low noise level using surface-enhanced hyper-Raman scattering. It is particularly remarkable that the hyper-Raman signal from one molecule can be stronger than the second harmonic generation from a complete plasmonic nanostructure, despite the considerable volume difference between both nano-objects. This fundamental observation stems from the different selection rules for both nonlinear optical processes.