Waveguide-enhanced Raman spectroscopy (WERS) is becoming a promising emerging technology, and this work is aiming at three major problems that general WERS system are currently facing. Traditionally, WERS focused on single-mode operation, but this study explores the impact of planar waveguide thickness on surface scattering losses with a dedicated theoretical model on scattering loss. By considering multimode cross-polarization signal collection, a new generalized Figure of Merit (FOM) that takes count of the scattering loss is built for planar WERS. It is revealed that multimode pump excitation yields much superior conversion efficiencies, despite the reduced surface intensity. As a by-product, a method for reducing scattering loss and hence strengthen WERS with extra interface in waveguide substrate is also illustrated.Waveguide grating couplers (WGCs), which offer high coupling efficiencies and relaxed alignment tolerances for WERS sensors, has been modelled with approximately treating the grating teeth as a special optical layer that is filled with the polarized dipole. The in-coupling of WGC is then calculated with out-scattering based on the reciprocal theorem, while the out-scattering is calculated with the theorem on dipole radiation at interfaces. The approximation extensively simplified the analytical difficulty for theoretical modelling. The accuracy of the model has been verified with a numerical model and the important physical mechanism that can be exploited for enhancing the Waveguide grating couplers has been revealed.Finally, a set of algorithms specifically for processing the signal from WERS that suffers from strong background and low resolution is proposed and implemented, showing good performance. The processing includes denoising, baseline removal and deconvolution. The novel multi-frame nonlocal means method has been proven more effective with simpler parameter settings than traditional methods. The asymmetric least square method has been strengthened by incorporating a radial basis function for more precise baseline removal while proving constraint to deconvolution to suppress the ringing artefact.