Numerical modelling with the Boundary Element Method is often hindered due to the excessive computational cost at higher frequencies. In that context, a suitable model order reduction scheme is proposed to alleviate the computational cost needed for a BE fast frequency sweep acoustic analysis. First, the method utilizes a kernel series expansion to remove the frequency dependency from the BEM system matrices. Subsequently, a Galerkin projection is deployed on the frequency independent matrices to reduce the size of the system. The basis employed in the Galerkin projection is assembled by the Krylov subspaces obtained through an Arnoldi procedure of the BEM system on a predefined frequency grid. The use of Krylov subspaces facilitates the definition of an error estimator that indicates the level of the error expected by the projection of the system. The efficiency of the model order reduction scheme is first assessed in terms of algorithmic efficiency and then validated on a test case.