Here is a recipe:

(define-public fplll
  (package
   (name "fplll")
   (version "4.0.1")
   (source (origin
            (method url-fetch)
            (uri (string-append
                  "http://perso.ens-lyon.fr/damien.stehle/fplll/libfplll-" 
version ".tar.gz"))
            (sha256 (base32
                     "122bpqdlikshhd7nmq0l5qfc0agyk7x21gvplv1l9hb77l8cy9rw"))))
   (build-system gnu-build-system)
   (inputs `(("gmp" ,gmp)
             ("mpfr" ,mpfr)))
   (synopsis "fplll, a library for LLL-reduction of euclidean lattices")
   (description
    "fplll LLL-reduces euclidean lattices. Since version 3, it can also 
solve the shortest vector problem.")
   (license "LGPLv2.1")
   (home-page "http://perso.ens-lyon.fr/damien.stehle/fplll/")))

It could go into multiprecision.scm, or get its own file.

Andreas