Massive Beamforniing systems in next-generation Centralized Radio Access Network architectures can substantially increase the requirements on fronthaul interfaces. This paper proposes a novel and efficient model to reduce the overhead generated by the transfer of beamforming coefficients between base station elements. This approach creates a compressed version for said coefficients, in the beamspace domain, using a clustering strategy over sets (windows) of most powerful beams. The method can be extended for any antenna configuration and also parameterized to adjust the trade-off between radio performance and the fronthaul load reduction. Simulations in typical mobile outdoor scenarios show that the proposed method can present Signal-to-Interference-plus-Noise-Ratio performance comparable to the usual dimensionality reduction approach, i.e., beam selection, while generating lower fronthaul overhead.