Let M denote an n-dimensional Riemannian manifold with structure group G. We assume that G acts transitively on the unit sphere in the tangent space and denote by $H \subset G$ the isotropy subgroup for some point in the sphere. We can show Let V_1 and V_2 be two G-Moduls whose restrictions to H are equivalent, then there exists an elliptic pseudo-differential operator from the sections of the bundle associated to V_1 into those for V_2. Consequently, to describe the space of elliptic operators naturally associated to the structure group G we need to determine the kernel R(G,H) of the restriction mapping R(G) -> R(H) of character rings. R(G,H) is a finitely generated ideal in R(G). Here are some examples: i) G = SO(2m), H = SO(2m-1), n=2m, R(G,H) = (1-\Lambda^1\pm\cdots- \Lambda^n-1+1, \Lambda^m_+-\Lambda^m_-). ii) G = Spin(2m), H = Spin(2m-1), R(G,H) = (\Sigma^+-\Sigma^-). iii) G = U(m), H = U(m-1), R(G,H) = (1-\Lambda^1,0\pm\cdots +(-1)^m \Lambda^m,0). We may call the operators corresponding to the generators fundamental. For the above examples this means that for oriented even-dimensional Riemannian manifolds the Euler operator and the signature operator are fundamental, for even-dimensional spin manifolds the Dirac operator is fundamental, and for almost complex manifolds the Cauchy Riemann operator is fundamental. It is interesting to apply this approach to some less well understood structure groups such as G = Sp(q)Sp(1), n=4q. We get elliptic operators for almost quaternionic manifolds and give new integralitiy and vanishing theorems in the case q odd. This method can also be used to study embeddability questions of manifolds into higher dimensional ones.