Considering viscous and thermal losses may be crucial for valid acoustic analysis. How much they affect such analysis depends on the properties of the propagation medium, the characteristic size of the fluid domain, and the acoustic wavelength. These losses take place mainly on the acoustic boundary layer, a region adjacent to the solid boundary. They must not be neglected in many real-world applications such as hearing aids, condenser and micro-electromechanical system microphones, and small-scale acoustic metamaterials. Different well-known acoustic models take them into account. Among those, the Full Linearized Navier-Stokes is the most robust, though computationally demanding to solve numerically. Attempting to do so with the Finite Element Method comes with the necessity of special meshing for the boundary layer and added degrees of freedom. Recent research efforts have focused on efficiently tackling this model with a Boundary Element Method formulation, showing well-conditioned systems’ matrices [1]. The present work extends the scope of the formulation to efficiently tackle multi-frequency problems through projection-based Model Order Reduction. During an offline stage, a projection basis is constructed via the Automatic Krylov subspace Recycling algorithm [2], which has shown to be efficient for systems with relatively clustered eigenvalues. Then, considering the non-affine parametric dependency of the Boundary Element systems on frequency, a frequency decoupling strategy built on Chebyshev polynomial approximation is employed. To minimize memory requirements, the projection is done immediately after each single system assembly necessary for the matrix polynomial approximation. Finally, during an online phase, the ensuing Reduced Order Model can be swiftly solved within the frequency range of interest. The performance of the numerical framework is assessed through numerical simulations of Helmholtz resonators problems with significant viscothermal dissipation. Results show the method achieves substantial speed-ups in comparison with its Full Order Model counterpart.