Random acceleration is a fundamental stochastic process encountered in many applications. In the one-dimensional version of the process a particle is randomly accelerated according to the Langevin equation $\ddot{x}(t) = \sqrt{2D} ξ(t)$, where $x(t)$ is the particle's coordinate, $ξ(t)$ is Gaussian white noise with zero mean, and $D$ is the particle velocity diffusion constant. Here we evaluate the $A\to 0$ tail of the distribution $P_n(A|L)$ of the functional $I[x(t)]=\int_0^{T} x^n(t) dt=A$, where $T$ is the first-passage time of the particle from a specified point $x=L$ to the origin, and $n\geq 0$. We employ the optimal fluctuation method akin to geometrical optics. Its crucial element is determination of the optimal path -- the most probable realization of the random acceleration process $x(t)$, conditioned on specified $A\to 0$, $n$ and $L$. This realization dominates the probability distribution $P_n(A|L)$. We show that the $A\to 0$ tail of this distribution has a universal essential singularity, $P_n(A\to 0|L) \sim \exp\left(-\frac{α_n L^{3n+2}}{DA^3}\right)$, where $α_n$ is an $n$-dependent number which we calculate analytically for $n=0,1$ and $2$ and numerically for other $n$. For $n=0$ our result agrees with the asymptotic of the previously found first-passage time distribution.