The southern Granite-Rhyolite province contains a comprehensive record of lithospheric evolution in North America. During the last decade, increased seismicity along with improved seismic monitoring installations in Oklahoma provided a rich catalog of local earthquakes. The source-receiver geometry of this dataset is well posed to illuminate the middle and lower crust through long offset recordings of the Pg phase. We present a 3-D P-wave velocity model for central and north Oklahoma developed through a non-standard processing scheme applied to local earthquake waveforms recorded from 2010-2017, focusing on the deeper crust. We employed common-mid-point sorting, stacking, and inversion of Pg-phases which resulted in a set of localized velocity-depth functions up to depths of 40 km. Using this methodology, we significantly increased the S/N ratio for far offset (~250 km) local earthquake waveforms which led to the increase in depth of investigation for our final 3-D velocity model. We find high velocity (> 7 km/s) lower crust throughout the investigated area which suggests a mafic lower crust. The high velocities support previously established models which state that the lower crust of the Granite-Rhyolite province was derived from melting of older crust. We further relate shallow and middle crustal velocity anomalies to other data sets such as gravimetric and magnetic anomalies, and the spatial distribution of earthquakes. We interpret the Nemaha Fault system as a deep-rooted discontinuity which separates two crustal domains. On the contrary, we do not find clear evidence for the existence of the Midcontinent rift (MCR) in northern Oklahoma.